1. Field of the Invention
The present invention relates to noise suppression and, more particularly, to noise suppression for multi-microphone sound pick-up systems.
2. Description of the Related Art
Suppressing interfere noise is still a major challenge for most communication devices involving a sound pick up system such as a microphone or a multi-microphone array. The multi-microphone array can selectively enhance sounds coming from certain directions while suppressing interferes coming from other directions. The pattern of the direction selection can be fixed or adaptive. Adaptive selection is more attractive because it intends to maximize SNR depending on the sound environment. However, because the relative low frequency range of audio applications, existing adaptation techniques are effective only for microphone array with large physical dimension. For applications where physical dimension is limited, such as the case in hearing aid applications, traditional adaptation using Finite-Impulse-Response (FIR) adaptive filtering techniques is not effective. As a result, most hearing aids that have directional processing can only give a fixed directional pattern which is effective in improving Signal-to-Noise Ratio (SNR) in some conditions but less effective in other conditions.
FIG. 1 shows a typical directional processing system in a two-microphone hearing aid. The two microphones pick-up sounds and convert them into electronic or digital signals. The output signal from the second microphone is delayed and subtracted from the output signal of the first microphone. The result is a signal with interference from certain directions being suppressed. In other words, the output signal is dependent on which directions the input signals come from. Therefore, the system is directional. The physical distance between the two microphones and the delay are two variables that control the characteristics of the directionality. For hearing aid applications, the physical distance is limited by the physical dimension of the hearing aid. The delay can be set in a delta-sigma analog-to-digital converter (A/D) or by use of an all-pass filter.
FIGS. 2(a)-2(c) illustrate polar patterns of a directional processing system corresponding to three different delay values. The term “polar pattern” has often been used to describe the characteristics of a directional processing system. The physical distance between the two microphones of the directional processing system is fixed. When a sound source is at 0 degrees, which is the direction along the axis of the two microphones and on the side of the front microphone, the directional processing system has a maximum output. When the sound source is away from 0 degrees, the output is reduced. The direction at which the output of the directional processing system has a maximum reduction is called directional null. Ideally, the directional null occurs at the direction of an unwanted noise source. The location of the directional null is related to the value of the delay. If the noise source is in the direction of 180 degrees, the delay should be set to a value so that the polar pattern is a cardioids with the directional null at 180 degrees (see FIG. 2(a)). If the noise source is in the direction of 115 degrees, the delay should be set to a value so that the polar pattern is a hyper-cardioid with the directional null at 115 degrees (see FIG. 2(b)). If the noise source is in the direction of 90 degrees, the delay should be set to a value so that the polar pattern is a bi-directional with the directional null at 90 degrees (see FIG. 2(c)). Ideally, the delay should be set in such a way that the null is placed in the direction of the dominant noise source so that the noise can be highly suppressed. If the direction of the noise source is known, the optima delay can be calculated as:delay=d/c*cos(180°−q),where d is distance of the two microphones, c is sound propagation speed, and q is direction angle in degree of the noise source.
One problem with conventional noise suppression approaches is that the direction of a noise source to be suppressed by the directional processing is often unknown. Conventionally, the estimating of the direction of a noise source is difficult because the frequency of audio sounds is relative low. The direction of the noise source is often merely a rough estimate from which a delay is then fixed to provide directional processing. In fact, most hearing aids currently available in the market merely set the delay to a fixed value so that directional processing has a fixed polar pattern for all conditions. Unfortunately, the noise suppression of such devices is often inadequate because the noise source is often at a direction other than that corresponding to the fixed delay.
Thus, there is a need for improved approaches to directional processing by adapting a directional null according to the direction of interfering noise source.